“This is a very flawed mental model. But what it really comes down to is, you could hit a tennis ball with a tennis racket at a certain force with the strings at a certain tension, and you could model that whole system. Right?
You could do it analytically with mathematics, and a physicist would want to start there and use closed form, partial differential equations. And you end up with what essentially becomes a solver that’s something that iterates, and you have to do numerical methods and use a lot of computing power. So this is simulation up to this point with electromagnetics. It’s the same kind of thing.
Or you can train a human to hit the tennis ball, and they practice a million times, and they become really good at tennis.
AI is essentially way back when, Marvin Minsky and others in the fifties were starting to realize that we could model the way human neurology works in software using this thing called a neural network. And it’s taking that thing and finding ways to train it so that it’s equivalent to training a tennis player to hit the ball. Now the tennis player is not doing math in their head. They’re not consciously playing with equations. Because if they were, they’d just stop and have to think the whole time.
This is where we’re at with things like electromagnetic simulations, why you need massive amounts of compute power. This is why a traditional field solver may take hours and hours to solve a simple electromagnetics structure, and a well-trained AI model can just recognize where the fields and waves will be. Just like a well-trained electrical engineer who’s been doing RF design for many, many years and has already visualized in their own mind can intuitively understand, oh, this structure here is going to have impedance discontinuity at this frequency. And I know because I’m looking at it, and I can see roughly the size of it.
That’s the difference we’re talking about. So, in a way, we’re modeling or replicating human intuition. I might be way off with that analog, but correct me. I think it’s a great analog.”
Watch Ben Jordan discuss AI and PCB layout with Sergiy Nesterenko in Trace Talks EP 7.